cross-crate intra-doc failures generate a broken link instead of stripping the link

[ben0x539]

Example: tracing-subscriber defines the MakeWriter trait with the make_writer method whose docs link to the Metadata type in tracing_core: https://docs.rs/tracing-subscriber/0.3.11/tracing_subscriber/fmt/writer/trait.MakeWriter.html#method.make_writer_for

tracing-appender has a type that implements MakeWriter and hence its docs reproduce the docs for the trait's methods: https://docs.rs/tracing-appender/0.2.2/tracing_appender/rolling/struct.RollingFileAppender.html#method.make_writer_for

The intra-doc link to Metadata didn't get resolved, and it just 404s: https://docs.rs/tracing-appender/0.2.2/tracing_appender/rolling/tracing_core::Metadata. tracing-core isn't a direct dependency of tracing-appender so I guess naming an item from tracing-core "in tracing-appender" doesn't work, even though the item is in scope where the doc comment physically exists.

Those crates live in a workspace at tokio-rs/tracing.

Here's a smaller example without a workspace: https://ben0x539.github.io/intradoclinkbreakage/im/target/doc/im/struct.Im.html, from https://github.com/ben0x539/intradoclinkbreakage.

idk if this is a known limitation, and I'm currently unable to think of the right search terms to find out. Seems unfortunate though.

Meta

 $ rustdoc --version --verbose
rustdoc 1.62.0-nightly (e7575f967 2022-04-14)
binary: rustdoc
commit-hash: e7575f9670f3c837def3d186ae09366c75c7632e
commit-date: 2022-04-14
host: x86_64-unknown-linux-gnu
release: 1.62.0-nightly
LLVM version: 14.0.0

[jyn514]

This is an intentional decision by cargo: rust-lang/cargo#8296 (comment)

[ben0x539]

Cargo decisions aside, emitting broken links sounds like a bug to me. Should we just strip the links in these cases where we cannot resolve them?

[jyn514]

@ben0x539 I think what's going on is #56922; we don't report the broken link because we don't think it's controlled by the same author as the original crate.

That said, it's weird we don't strip the link, I think we do normally for failures in the same crate. Would be good to test that and see why it differs.